Sensors 2015, 15, 22705-22723; doi:10.3390/s150922705OPEN ACCESS

sensorsISSN 1424-8220

www.mdpi.com/journal/sensors

Article

A Simplified, Light Emitting Diode (LED) Based, ModularSystem to be Used for the Rapid Evaluation of Fruit andVegetable Quality: Development and Validation on Dye SolutionsRaffaele Civelli, Valentina Giovenzana, Roberto Beghi *, Ezio Naldi, Riccardo Guidetti andRoberto Oberti

Department of Agricultural and Environmental Sciences, Università degli Studi di Milano,Via Celoria 2, Milano 20133, Italy; E-Mails: raffaele.civelli@unimi.it (R.C.);valentina.giovenzana@unimi.it (V.G.); ezio.naldi@unimi.it (E.N.); riccardo.guidetti@unimi.it (R.G.);roberto.oberti@unimi.it (R.O.)

* Author to whom correspondence should be addressed; E-Mail: roberto.beghi@unimi.it;Tel.: +39-2-50316843; Fax: +39-2-50316911.

Academic Editor: Vittorio M. N. Passaro

Received: 16 June 2015 / Accepted: 31 August 2015 / Published: 8 September 2015

Abstract: NIR spectroscopy has proven to be one of the most efficient and ready to transfertools to monitor product’s quality. Portable VIS/NIR instruments are particularly versatileand suitable for field use to monitor the ripening process or quality parameters. The aimof this work is to develop and evaluate a new simplified optoelectronic system for potentialmeasurements on fruit and vegetables directly in the field. The development, characterizationand validation of an operative prototype is discussed. LED technology was chosen for thedesign, and spectral acquisition at four specific wavelengths (630, 690, 750 and 850 nm) wasproposed. Nevertheless, attention was given to the modularity and versatility of the system.Indeed, the possibility to change the light sources module with other wavelengths allows oneto adapt the use of the same device for different foreseeable applications and objectives, e.g.,ripeness evaluation, detection of particular diseases and disorders, chemical and physicalproperty prediction, shelf life analysis, as well as for different natures of products (berry,leaf or liquid). Validation tests on blue dye water solutions have shown the capability ofthe system of discriminating low levels of reflectance, with a repeatability characterized bya standard deviation proportional to the measured intensity and in general limited to 2%–4%.

Sensors 2015, 15 22706

Keywords: portable optical device; non-destructive analysis; reflectance; ripening; fruitand vegetable

1. Introduction

The study of non-destructive methods and the design of new devices for monitoring a large number ofsamples in a short time and allowing a more comprehensive overview of ripening is an ongoing researchsubject [1–3].

To this aim, currently visible near-infrared (VIS/NIR) and near-infrared spectroscopy are techniqueswidely applied in the food sector [4]. A review of the literature reveals that NIR techniques (VIS/NIR andNIR) have been applied to a wide range of agri-food applications. The feasibility of NIR spectroscopyto measure quality attributes of fruit and vegetables has been shown for many products [2]. Datacomplexity arising from NIRs requires specific statistical analyses and qualified operators. Furthermore,nowadays, the available devices are quite expensive and therefore not suitable for small-scale producers.

During fruit ripening, biochemical changes occur not just at the skin level, but also in the pulp. NIRsanalyses allow one to reach the inner layers of the sample when appropriate wavebands are used. Forthese reasons, NIR spectroscopy resulted in being a suitable technique to evaluate ripeness in the orchardand post-harvest quality characteristics of fruits.

To this aim, three main types of NIR devices can be identified: (i) laboratory instruments forapplications in research centers or in industry laboratories; (ii) sorting and grading devices designedspecifically for the fruit and vegetable industries, e.g., in warehouses; (iii) portable devices for usedirectly in the field. Table 1 shows the main differences between the three types of NIR devices.

Table 1. Characteristics of the three main categories of NIR devices.

Application Area Flexibility of Use ApplicabilityMeasurementAccuracy and

ReproducibilityCost

Laboratorydevices Research/Industry

Adaptable todifferent matrices Fixed system Optimal Average/high

Sorting andgrading Industry

Specific categoriesof products Fixed system Fair Average/high

Portabledevices Also in field

Dedicated forindividual products Portable/handheld Fair Average

Nowadays, a wide selection of spectroscopy devices at different complexity levels is available, andthere are about 60 NIR spectrometer manufacturers around the globe [5].

For every device category, calibration models to be used in practice need to be based on largedatasets, encompassing several orchards, cultivars, growing conditions, seasons and device operatingconditions (e.g., temperature, environmental lighting). All of these factors have to be incorporatedwithin appropriate preprocessing methods or to be compensated [2,3], aiming at optimizing predictionrobustness. Sample presentation to the instrument is also a crucial step in NIR analyses. Specialized

Sensors 2015, 15 22707

sampling probes, liquid cells and accessories have been manufactured to meet measurement demandsaccording to how and where they will be performed (e.g., in a laboratory, at a production line or in thefield) [3,6].

The availability of handheld spectrophotometers has opened up the possibility to use them in theorchard for monitoring fruit maturity. Scientific literature reports some studies on applications ofportable NIRs to this aim. In most cases, limited prediction accuracy was obtained. For example, byusing a portable VIS/NIR instrumentation, Antonucci et al. [7] predicted the titratable acidity on twocultivars of mandarin getting R2 a range of 0.66 – 0.77, while for the soluble solids content, a R2 rangeof 0.71 – 0.72 on apricot. Camps and Christen [8] estimated the soluble solids content, the total acidityand the firmness achieving R2 in cross-validation in ranges of 0.77 – 0.92 0.53 – 0.94 and 0.72 – 0.85,respectively. Beghi et al. [9] on golden delicious apples, determined R2 the soluble solid content,chlorophyll, titratable acidity, flesh firmness,total phenols, carotenoids and ascorbic acid, equal to 0.72,0.86, 0.52, 0.44, 0.09, 0.77 and 0.50, respectively.

This might be due to factors affecting model robustness, such as temperature fluctuations,uncontrolled lighting conditions, the limited wavelength range or the fact that the devices were stillat a prototype stage. The development of portable devices suitable for field use is more complexthan laboratory applications, due to the uncontrolled environmental conditions, such as ambient light,fluctuating temperatures, power source, illumination, but also the need of instrumentation with alower technological complexity. All of these phenomena should be addressed by appropriate dataprocessing [2].

Portable VIS/NIR instruments were tested in controlled laboratory conditions by Antonucci et al. [7]for the evaluation of mandarin maturity status and by Camps and Christen [8] for assessing apricotquality. Under more difficult uncontrolled field conditions, instead, a portable VIS/NIR device wastested in order to estimate apples’ nutraceutical properties [9], to evaluate grape quality parameters [10],to assess the ripeness of red-pigmented fruits [11] and to predict blueberry ripeness [12].

Research and innovations have enabled NIRs devices to further decrease their physical size whileincreasing complexity and the amount/size of collected data. Therefore, new NIRs instrumentation tendsto be more compact and portable [3,5] opening new possibilities for field use.

Few portable commercial spectrophotometers, for example the Fruit Tester 20 (FANTEC, Kosai-city,Japan) [13], the Jaz Modular Optical Sensing Suite (Ocean Optics Inc., Dunedin, FL, USA), and theQS_300 (Unitec SpA, Lugo, Italy), have been proposed for different food applications.

For each device category (Table 1), the development of calibration models and the extraction ofuseful information contained in spectral data rely on multivariate analysis. Chemometrics, in fact, isa crucial part of NIR spectroscopy applications. Indeed, these measurement techniques must always becomplemented with chemometric analysis to enable the extraction of useful information present in thespectra, by separating it both from useless information or from spectral noise [2,14–16].

Furthermore, the use of portable NIR devices requires skilled operators, able to process complex datain order to extract useful information and to build ripening prediction models [4,16]. Therefore, in orderto support farmers or small-scale producers, simplified, easy to use and low-cost devices for real-timemeasurements in field are desirable.

Sensors 2015, 15 22708

To reach this goal, considerable efforts have been recently directed towards developing and evaluatingdifferent procedures for an objective identification of a few spectral variables containing most of theuseful information and to reject redundant or useless variables [17]. In view of simplified optical systems,different approaches for variables selection were applied in previous studies conducted by the authors onfresh-cut Valerianella [18], on wine grapes [19] and on blueberries [20].

Only a few examples of simplified optical systems have been reported in the literature. For example,the University of Bologna [21] patented a new simplified NIR device in a two-fold version (DA-Meterfor apples, and Kiwi-Meter for kiwi fruits, where DA refers to a maturity index introduced by the authorsbased on the difference in absorbance between specific wavelengths). These devices evaluate the fruitmaturity stage through indices based on absorbance differences among specific wavelengths.

The aim of this work was to develop a simplified four-wavelength optical system, based on LEDtechnology, potentially to be used for the rapid monitoring of fruit and vegetable parameters directlyin field (e.g., ripeness evaluation, disease and disorder detection, chemical and physical propertyprediction, freshness level or shelf life analysis). In order to assess the performance of the prototype,the characteristic curve and the operating point were investigated, and tests on standard solutionswere performed.

2. Experimental Section

The design of the measurement system pursues different objectives: low cost, low complexity (thanksto the use of a microcontroller), modularity, tuning possibilities and versatility, i.e., the capability toadapt to the different needs required by a specific application. The system is based on the acquisition ofthe spectral reflectance at wavelengths of interest. Light emitting diode (LED) technology was chosenas the light source [3], so as to achieve the capability to individually adjust the light emission intensityfor each measurement channel.

2.1. System Hardware

Figure 1 shows a diagram of the instrument with the component units in relationship to theirfunctionalities. In particular, as shown by the figure, the system is composed of:

1. the control and processing unit;2. the interface unit;3. the analog-to-digital and the digital-to-analog converters;4. the LEDs and optical filter modules;5. the photodiodes;6. the eight-arm optical fiber;7. the power supply system.

2.1.1. The Control and Processing Unit

This is the main board of the device, equipped with a PICTM microcontroller (ProgrammableIntegrated Controller, PIC18F series, Microchip Technology Inc., Chandler, AZ, USA), (1 in Figure 1).

Sensors 2015, 15 22709

When the instrument is turned on, the microcontroller automatically runs a preloaded firmware forthe initialization, and manages the inputs entered by the user. Based on these, it checks the stateof the system and sends signals to the other units in order to operate. The main program waswritten using a development tool based on a C language compiler and then stored on the memory ofthe microcontroller.

Figure 1. Block diagram of the main components of the prototype.

2.1.2. The Interface Unit

The user can control the system through a four-button keyboard connected to the main board (2 inFigure 1), which represents the input interface. By using the keyboard, the operator can control differenttasks, e.g., he can start the calibration procedure, make a new measurement or delete the last one.

The output interface is an alphanumeric four-line LCD display (2 in Figure 1). The acquired dataand other messages are displayed in real time, giving the user instantaneous feedback to check if theinstrument is working correctly. At the same time, the data are sent to a palmar tablet or a notebook PCthrough a wireless Bluetooth card. In this way, the work session can be followed on a PC monitor, anddata can be saved for any post-processing.

2.1.3. The Analog-to-Digital and the Digital-to-Analog Converters

The communication between a digital component, such as the microcontroller in the control unit,and the actuators (LEDs) or the transducers (photodiodes) requires the conversion of the voltage/currentsignals from digital to analog form and vice versa.

Sensors 2015, 15 22710

This task is accomplished by the 8-bit digital-to-analog converter (DAC; 3 in Figure 1), for themodulation of the LEDs’ driving current and the 16-bit (1 bit for the number sign, 15 bits for theencoding of the numerical value) analog-to-digital converter (ADC; 3 in Figure 1), which convertsthe analog inputs measured by the photodiodes in digital form. Both of these devices use the I2CTM

(Inter-Integrated Circuit) serial interface as the communication protocol.

2.1.4. The LEDs and Optical Filters Modules

The instrument can be equipped with up to four independent light sources, whose luminousintensities can be individually adjusted and set. The requirement for accurately driving the luminousradiation within a wide range of values has been met using LEDs as light sources. In this way, accordingto the specific application, the system can emit a custom 4-wavelength spectrum, e.g., the analysis ofred grape needs a high illumination intensity close to the anthocyanins absorption peak (around 520 nm),while it requires a low intensity near the third-overtone of the OH bond (around 740 nm).

The choice to limit the number of the measurement channels (channel = LED + filter + photodiode)to four has been established to reduce the complexity of the overall optical system, since this must besimple and handheld.

The current configuration of the device is based on the spectral bands selected in previousstudies [19]. The LEDs actually used were chosen at the wavelengths that most closely matched thetheoretical choice, according to the availability on the market. Therefore, the wavelengths used for thesystem are: 630 and 690 nm, near the chlorophyll absorption peak, 750 nm, near the third-overtoneof OH bond stretching, and 850 nm, arbitrarily elected among the uninformative wavelengths fornormalization purposes.

All four LEDs mounted in the prototype come from the same LED series (ELJ series, RoithnerLasertechnik GmbH, Vienna, Austria). These are high power LEDs designed for several opticalapplications, including use in measurement systems. The LED technology is particularly appropriateto be adopted in the present device. This is explained point by point by the following arguments, withreference to the technical data summarized in Table 2.

- The ELJ family is based on semiconductor materials (e.g., aluminum-gallium-indium-phosphideor gallium-aluminum-arsenide) that ensure highly accurate and well-defined wavelengths forthe emitted radiations in the VIS/NIR range, with a narrow bandwidth. The peak wavelengthtolerances do not exceed ±10 nm, with bandwidths ranging between 20 and 40 nm. Any possibleshift of the emission wavelength due to the drive current changing is minimal (in the order of1 – 3 nanometers). This effect is further compensated along every measuring channel by theintroduction of a dedicated filter, which will be discussed later.

- The light intensity emitted by an LED is easily modulated by using the drive current with a fairlylinear response (the study of the linearity of the overall response for the optical channel, i.e., thecharacteristic curve, is among the objective of this work). This aspect, together with the reduceswitching time, makes an LED particularly suitable to be controlled by the current signals comingfrom the microcontroller. Furthermore, the individual control of each LED makes it possible togenerate a customizable 4-wavelength spectrum of emission, as stated above. Thanks to the use

Sensors 2015, 15 22711

of LED technology, the channels can be independently managed, and the measure settings can beindividually optimized.

- The light is emitted by an LED according to a narrow optical cone, which simplifies its collimationtowards the optical fiber and reduces any possible crosstalk effects between the different channels.In particular, the LEDs of the ELJ collection are provided in a metal case with a plastic lens, whichproduces a viewing angle between 15◦ and 20◦ according to the model.

- The ELJ series is characterized by a good efficiency, which means high values for radiant powerand intensity versus a relatively small dissipation power. Efficiency is useful with the intent ofbattery saving for applications in the field [2]. Typical values of the radiant intensities and radiantpowers for a drive current at zero wavelength shift are reported in Table 2, and so the maximalvalues of the power dissipation.

- Finally, the compact sizes and the low weight of LEDs give an additional argument to choose thistechnology for the development of a handheld and portable system.

Table 2. Technical data for each measurement channel LED (Roithner Lasertechnik GmbH),filter (Edmund Optics Inc.) and photodiode (Roithner Lasertechnik GmbH).

ItemLED Model

ELJ-630-628 ELJ-690-629 ELJ-750-629 ELJ-850-629

Semiconductor material - AllnGaP AlGaAs AlGaAs AlGaAsDrive current at zero wavelength-shift I0 mA 350 350 350 350

Max drive current IM mA 1000 700 1200 1200

Max power dissipation mW 3000 3200 4000 4000

Peak wavelength a nm 630± 5 690± 10 755± 10 850± 10

FWHMb Spectral bandwidth at 50% a nm 20 23 30 40

Viewing angle a deg 15 17 20 20

Radiant power a mW 95 55 80 100

Radiant intensity a mW/sr 830 380 650 2000

Switching time a ns 60 45 60 20

ItemFilter Model

#65-651 #65-660 #67-775 #67-785

Center wavelength nm 632± 2 694± 2 750 +3/−0 850 +3/−0

FWHM b nm 10± 2 10± 2 10± 2 10± 2

Minimum transmission % ≥45 ≥50 ≥50 ≥50Blocking wavelength range nm 200–1200 200–1200 200–10 000 200–10 000

Item Photodiode Model

Spectral range c nm 400–1100Maximum spectral responsivity Smax

c mV/nW 60

Wavelength at Smaxc nm 800

Viewing angle c deg ±50

a Typical values at 20 ◦C and for Drive current = I0, i.e., drive current at zero wavelength shift; b fullwidth-half maximum; c typical values at 25 ◦C and for supply voltage VS = ±15V.

Sensors 2015, 15 22712

Four optical filters (#65- and #67-series, by Edmund Optics Inc., Barrington, NJ, USA), one foreach LED, are positioned at the end of the optical fiber input arms (see Section 2.1.6), just beforethe photodiodes. Each filter was chosen with a center wavelength as close as possible to the peakemission wavelength of the corresponding LED, according to the availability on the market (seeTable 2). The #65- and#67-series have a narrow and well-defined spectral bandwidth of (10± 2) nmand ensure a transmission coefficient ≥45% , while filtering all other wavelengths in the visible andnear-infrared range. The introduction of the filters cuts down the wide-range environmental noise duringthe acquisition, resulting in an improvement of the signal-to-noise ratio.

The LEDs and the relative filters represent the module unit of the instrument (4 in Figure 1), whichcan be changed according to the specific application (see Section 2.2).

2.1.5. The Photodiodes

Each measurement channel is equipped with a silicon-based photodiode (IQ800L, RoithnerLasertechnik GmbH, Vienna, Austria) operating in the spectral range 400 – 1100 nm, with apeakresponsivity at 800 nm (see Table 2). Although the IQ800L is already provided with an integratedlow noise JFET-amplifier (junction gate field-effect transistor amplifier), an external dedicated stage ofelectronic filtering and amplification was designed for every photodiode in order to further improve thesignal-to-noise ratio, before the analog-to-digital conversion (5 in Figure 1).

2.1.6. The Eight-Arm Optical Fiber

A customized optical fiber was specifically designed for the application in order to optimizethe radiation transmission and the optical couplings, and an industry leader was commissioned forits realization.

This consists of a bundle of 32 multimode step index fibers with a low OH fused silica core andglass cladding (Fort Fibre Ottiche S.r.l., Curno, Italy), with a diameter of 600 µm and a length of 30 cmarranged in 8 independent arms in protective plastic sheaths, of 4 fibers each (see the pictures in Figure 2).The arms are free at one end, each one with a standard SMA connector (subminiature version A coaxialconnector), while at the other end, they all converge together in a single metallic probe where the fibersare mixed. This is the element that is put in contact with the sample during a measurement. Fourarms (output arms OUT; 6 in Figure 1) are connected to the LEDs, receive the emitted radiation andtransmit it towards the probe and then to the sample. Conversely, the other four arms (input arms IN;6 in Figure 1) collect the radiation coming back from the sample and transfer it to the four photodiodesfor the measurement.

2.2. Modularity

The prototype of the simplified optical device is designed with particular attention to versatilityand modularity concepts. It is desirable to have the possibility to adapt the simplified opticaldevice for different objectives, applications and different kinds of sample matrices, while keeping itsmain architecture.

Sensors 2015, 15 22713

This can be obtained by selecting the emission spectrum at specific wavebands by an appropriatechoice of the LEDs. Envisaged applications for this kind of instrument would be, for example, theassessment of the main technological and phenolic parameters of grapes for wine production [10] orthe early detection of disease infection (e.g., Botrytis cinerea or powdery mildew) on leaves [22,23]or berries.

(a) (b)

Figure 2. The customized quartz optical fiber: (a) front and (b) lateral view.

2.3. How the System Works

Since the very beginning of the development of the device, the approach of sequential (multiplexed)measurements was preferred over the simultaneous one. This was mainly due to both optical andelectronics reasons. From the optical point of view, the possibility to customize the emission spectrumby changing LEDs and filters (i.e., modularity concept) makes potentially possible applications basedon the investigation of very close wavebands, in particular to estimate the values of derivative spectra.The sequential design is therefore mandatory to mitigate the cross-talk effect due to the partial overlapof the emission/measuring spectral curves, which instead would occur with a simultaneous acquisition.Regarding the electronics aspects, the simultaneous activation of all channel LEDs was avoided for notincurring current load peaks in the system that could cause electronic inter-channels interferences, and,more in general, for power saving reasons in a portable, battery-powered device.

According to the acquisition procedure, when the user presses the dedicated button (2 in Figure 1),the microcontroller runs a memory pre-loaded routine, which controls the individual powering of theLEDs, according to a specific sequence. With reference to Figure 1, the acquisition routine can beschematically described as follows:

1. The microcontroller (1) sends the driving signal to power one of the LEDs.2. As soon as the individual LED turns on, the emitted light is brought to the sample through the

corresponding output arm of the optical fiber. The radiation interacts with the sample matrix, andthe reflected back-scattering light is collected by the input arm (6).

Sensors 2015, 15 22714

3. At the same time, the microcontroller enables the detection on the corresponding photodiode (5) foran instantaneous acquisition. The output signal of the photodiode is than amplified and digitizedby the ADC (3) and finally acquired by the control and processing unit of the microcontroller.

4. The measurement at one wavelength is obtained as an average of 5 acquisition cycles as repeatedin sequence by the processor of the microcontroller. If the deviation among repetitions is higherthan 3% of the average, the measurement is rejected and the sequence repeated.

5. The same steps (1)–(4) are sequentially repeated for all of the channels.6. Once this procedure is completed for all of the LEDs, the data are ready to be shown on the

display (2). In the same way, they are sent to the Bluetooth card (2) and then to the computer. Theoverall acquisition procedure takes less than 1500ms, after which the device is ready for a newacquisition. A single acquisition is in practice almost instantaneous, considering that the deviceis intended for manual use in static applications.

2.4. Setting Parameters and Operating Characteristic Curves

The prototype is designed to perform a direct measurement of the reflected light intensity at specificwavelengths, expressed in an arbitrary unit coming from the electronics of the system (in a range between0 and 32767).

The quality of the direct measurement is a prerequisite to obtain the correct estimation of thereflectance, i.e., the ratio at a given wavelength between the intensity of the radiation reflected bythe sample and the intensity of the illuminating radiation. The reflectance is the real significantvariable, which allows comparisons between acquisitions made at different times and under differentenvironmental conditions. The choice of the device setting parameters affecting the measurement outputbecomes therefore crucial.

The setting parameters should be chosen in order to establish a proper operating point along thecharacteristic curve of the system for each channel, both by ensuring the linearity of the response and byexpressing an adequate sensitivity with respect to the specific application. In particular, measuring thesame sample under the same conditions, the instrumental parameters, which can be set and that influencethe detected light intensity on a certain channel, are two:

- The stimulation level of the LED drive current, Stimulus , which modulates the intensity of theincident light and, consequently, of the reflected light, as well;

- The amplification factor of the signal detected by the photodiode, Gain, which defines thenumerical value actually read within the available measurement scale.

Following this, the response of the device as a function of the Stimulus and Gain parameters, i.e.,the characteristic curve of every channel of the system, was investigated, in view of the definitionof a calibration procedure for setting an appropriate operating point, optimal for the application ofthe device.

A special test software has been implemented and loaded into the microcontroller to automaticallyexplore the totality of the operating parameters space. With this configuration of the device, it waspossible to evaluate the response for all of the admissible (Stimulus ,Gain) couples (with integer valuesof Stimulus in the range 0 < Stimulus < 255 and Gain = 1, 2, 4, 8). The tests were repeated

Sensors 2015, 15 22715

on the four available standard references (Spectralon R© Diffuse Reflectance Standards, Labsphere Inc.,North Sutton, NH, USA), for reflectance coefficients R = 0.99, 0.50, 0.25, 0.02.

2.5. Identification of a Simple Calibration Procedure for Defining the Operating Set-Point

In order to simplify the use of the instrument, the full capability of the setting parameters(Stimulus ,Gain) is not given to the user. These operating parameters have in fact to be properlyset at the beginning of every working session, when the instrument is switched on. However, theiridentification, i.e., the setting of an operating point along the characteristic curve for each channel, isapplication specific and in general not immediate.

For example, when measuring a dark-colored fruit or cultivar, like red grapes or blueberries, a highsensitivity through all of the visible range will be in general required. On the contrary, for a light-coloredfruit, e.g., white grape cultivars, a high sensitivity in the red spectral range will be necessary (due tochlorophyll absorption in this range), while a low sensitivity at other visible wavelengths will be ingeneral necessary; or again, when measuring green leaves, sensitivity needs to be optimized for differentwavelengths: a partially low sensitivity will be required in the green range due to relatively high leafreflectance; while a high sensitivity in the red band and low sensitivity in the near-infrared range willbe necessary. Starting from the study of the characteristic curves, a simple calibration procedure, basedon acquisitions made on standard reflectance references was defined (see Section 3.2) to properly set theoperating point for each channel.

2.6. Validation Tests on Custom Samples

The device was then tested on three custom standard samples, consisting of distilled water solutionsof blue food dye (Brilliant Blue R, CAS Number 6104-59-2, Sigma-Aldrich Co. LLC., St. Louis, MO,USA) at concentration of 0.5% (w/v), 1% (w/v) and 2% (w/v), respectively.

An acquisition setup for this kind of liquid sample was defined. According to this, a solution sampleis inserted in a quartz 5mm path cuvette (100-QS, Hellma GmbH & Co., Mullheim, Germany), whichis then placed into a custom holder made of Teflon, i.e., from a commercial-grade PTFE bar, externallycovered with black cladding. A hole on the holder allows one to insert the optical fiber probe, guiding itto face the cuvette window, into a proper fixed position, which ensures repeatable measuring conditions.The cuvette containing the dye solution is therefore embedded into a diffuse reflecting material. Asa result, the detected photons passed through the sample before and after being reflected in the directionof the detection fibers, without being absorbed by the sample. As the sample holder is fixed for allmeasurements, the dye concentration is the only variable property on the sample side.

Two kinds of tests were conducted.First, the same explorative configuration presented in Section 2.4 was applied, and the characteristic

curves of the system were built, by measuring the three standard solutions while changing the Stimulus

parameter at the fixed Gain value, previously determined (Gain = 1). These experiments aimed toassess the capability of the device of measuring differences in reflectance at a very low level of intensity(i.e., at very high absorption). These conditions are expected, for example, when measuring dark fruitspecies, e.g., red grapes or blueberry, in the channels at 630 and 690 nm.

Sensors 2015, 15 22716

Secondly, the calibrated configuration of the instrument, ready for the operation, was validated bysimply iterating five acquisitions on the standard solutions and evaluating the measurement repeatability.As described above in Section 2.5, each of the four channels followed an independent calibration witha dedicated reference.

3. Results and Discussion

3.1. Operating Characteristic Curves

Each of the four channels has proven to have a similar characteristic curve. As an illustrative example,Figure 3 shows the output intensity obtained at the channel λ = 750 nm when measuring a standardreference target with a reflectance R = 0.25, respectively, while the Stimulus was changed at fixedGain values (Figure 3a) and while the Gain was changed at fixed Stimulus values (Figure 3b).

Stimulus (a.u.)5 10 15 20 25 30 35 40

Inte

nsity

(a.

u.)

#104

0

0.5

1

1.5

2

2.5

3

6 = 750 nm - Tests on std. ref. R = 0.25

G = 1G = 2G = 4G = 8

(a)G (unit)

1 2 3 4 5 6 7 8

Inte

nsity

(a.

u.)

#104

0

0.5

1

1.5

2

2.5

3

Curves inStimulus param.

6 = 750 nm - Tests on std. ref. R = 0.25

(b)

Figure 3. Acquisitions on channel λ = 750 nm, using reference R = 0.25: (a) outputIntensity vs. Stimulus for the four curves in the Gain (G) parameter; (b) output Intensityvs. Gain (G) in the Stimulus parameter.

A similar behavior was obtained for all four channels. In this regard, the measurement raw filescoming from these tests are accessible as supplementary materials (refer to Appendix A–D).

Figure 3a represents the output intensity measured as a function of the Stimulus variable, for the fourlevels of Gain that the system allows to set. Only the significant portion of the graphs is displayed, i.e.,below Stimulus values causing saturation of the output signal. The characteristic curve of the system’soutput intensity as a function of the Stimulus variable clearly exhibits a highly linear behavior, regardlessof the Gain value set. As expected, higher values of Gain correspond to an increase in the slope of thelinear curve.

The linearity of the response with the change of the Stimulus variable is then assured for every choiceof the Gain parameter and, at least in theory, of the operating point position along the curves betweenzero and saturation.

Sensors 2015, 15 22717

Nevertheless, saturation is reached very quickly, and actually, just a small portion of the availablerange 0 < Stimulus < 255 is used. The highest slope curves (high Gain values), combined withhighly reflective standard references could lead to the risk of reaching saturation even at the very firstStimulus values.

The phenomenon is pronounced in measurements on a standard reference with a higher reflectance,and it is in general confirmed by the other tests (refer to Appendix A–D). This is a common behavior inall four channels due to the optical high efficiency of the system.

In practice, Gain = 1 is the only acceptable value for all four channels. It corresponds to the smallestslope curve, developing in the wider range of Stimulus values before saturation is reached.

Figure 3b leads to conclusions consistent with the previous ones. This plot shows the same acquisition(λ = 750 nm and standard reference R = 0.25) representing the output intensity measured as a functionof the Gain variable, for the 256 curves parameterized in Stimulus . In this case, only the curvescorresponding to the first values of the Stimulus parameter are distinguishable. All of the other curvescollapse on the horizontal saturation line (Intensity = 32 767 a.u.) as the Stimulus parameter increases;they are therefore overlapping. The linearity of the response with the change of the Gain variable is verygood, clearly for the curves not deformed by saturation effects (for a low value of Stimulus). A singleacceptable value Gain = 1 for all of the measurement channels is confirmed by this results, being theone providing the greatest number of significant Stimulus values.

Concerning the comparison among the different channels, the system shows the same kind of responsefor all four wavelengths (Appendix A–D). However, parameter values change in general depending onthe specific channel. This is due to the LEDs and photodiode features that change depending on theoperating wavelength, despite these components all coming from the same collections, as declared bythe manufacturers. This aspect is taken into account by simply making an independent calibration onevery channel using different standard references. Furthermore, this optimizes the behavior of eachchannel also depending on the spectral characteristics of the sample to be measured. While the intensitymeasures of reflected light on the four wavelengths are not comparable even in the same acquisition,comparability remains unchanged with regard to the reflectance measures obtained in post-processing.

3.2. Calibration

The calibration procedure was designed on the basis of the considerations about the characteristiccurves of the system, as a function of the Gain and Stimulus parameters.

Regarding the amplification Gain parameter, the above considerations (see Section 2.4) show thatG = 1 is the optimal value, independent of the sample and the application. The Gain parameter wastherefore determined una tantum and preset at this fixed value.

On the other hand, for the Stimulus parameter, a dedicated algorithm implemented in themicrocontroller automatically searches and sets the optimum for each channel, every time the instrumentis switched on.

According to this algorithm, an initial acquisition on a standard reference chosen by the user mustbe conducted. As schematically shown by the diagram in Figure 4a, the lowest level of Stimulus thatproduces a detected light intensity above an appropriate threshold is determined through the successive

Sensors 2015, 15 22718

incremental Stimulus signal. As a detected intensity threshold, the half value of the full availablescale (32 768/2 a.u. = 16 384 a.u.) was chosen. If calibrating with an appropriate standard reference,this threshold value avoids the risk of saturation during a working session, while ensuring a sufficientresolution to capture the dynamics of all expected samples.

Stimulus + 1

YES

NO

Set Stimulus

CALIBRATION ALGORITHM

I/O

USER

Chose Channel

Chose Std Ref

Chose Dark Std Ref I/O

USER

READY TO ACQUIRE

Th = Threshold

Initial Stimulus = 1

Intensity acquisition

Intensity acquisition

Intensity > Th

Intensity acquisition

(a)Stimulus (a.u.)

5 10 15 20 25 30 35 40In

tens

ity (

a.u.

)

#104

0

0.5

1

1.5

2

2.5

3

6 = 750 nm, G = 1 - Calibration procedure

R = 0.99R = 0.50R = 0.25R = 0.02ThresholdOper. Point

(b)

Figure 4. (a) Block diagram for the calibration procedure along a generic channel; (b) resultsof calibration on channel λ = 750 nm, using standard references R = 0.99, 0.50, 0.25, 0.02.

The value of Stimulus , which is defined in the calibration for each channel, is then used to powerthe corresponding LED during the whole working session, until the instrument is switched off. In thisway, the user has only an indirect effect on setting the optimal Stimulus for his/her specific application(i.e., with the choice of the reference used in the calibration). Once the Stimulus level is determined,an acquisition of a dark reference (R = 0.02) concludes the calibration procedure.

Figure 4b presents the result obtained with this procedure on channel λ = 750 nm for a calibrationwith the standard references R = 0.99, 0.50, 0.25, 0.02. The sequential increments of the detected lightintensity produced by the increasing Stimulus signal are shown. The algorithm stops as soon as thethreshold value is exceeded, and the last value of Stimulus is thus set to control the corresponding LEDduring all working sessions.

The plots of Figure 4b also highlight how the slope of the line changes with the standard referenceused in the calibration, which results in different values determined by the algorithm for the Stimulus

parameter. As an illustrative example, the attempt to calibrate with the dark standard reference(R = 0.02) is also shown. As expected, this resulted in the impossibility to reach the threshold value bythe measured intensity.

These results show how the only thing the operator has to pay attention to is the choice of thestandard reference when calibrating. According to this procedure, the reference should provide a spectralreflectance as similar as possible to the ones of the samples to be measured, ensuring a proper positioningof the operation point along the characteristic curve for the specific application.

The same calibration procedure has to be repeated for each channel, i.e., for each LED, by usingappropriate standard references with respect to the spectral characteristic of the samples, in order tooptimize the acquisition along every channel.

Sensors 2015, 15 22719

3.3. Response and Validation Tests on Standard Solutions

Figure 5 presents the results coming from the tests on the aqueous solutions of blue dye atconcentrations of 0.5% (w/v), 1% (w/v) and 2% (w/v), respectively. The curves of the output intensitymeasured as a function of Stimulus variable for the three samples are shown together in the same plot(the Gain parameter is now set at the fixed value G = 1, previously determined). For the sake of clarity,only the results for the two visible channels are presented (λ = 630 nm at the left and λ = 690 nmat the right), since these are the most critical spectral bands for dark samples and the ones that wewanted to investigate. Data referring to other channels are accessible as supplementary material, withthe measurement files in Appendix E–G.

Stimulus (a.u.)50 100 150 200 250

Inte

nsity

(a.

u.)

#104

0

0.5

1

1.5

2

2.5

3

6 = 630 nm, G = 1 - Tests on std. solutions

2% (w/v)1% (w/v)0.5% (w/v)ThresholdOper. Point

(a)Stimulus (a.u.)

50 100 150 200 250

Inte

nsity

(a.

u.)

#104

0

0.5

1

1.5

2

2.5

3

6 = 690 nm, G = 1 - Tests on std. solutions

2% (w/v)1% (w/v)0.5% (w/v)ThresholdOper. Point

(b)

Figure 5. Output Intensity vs. Stimulus with the Gain parameter G = 1 for acquisitionson aqueous solutions of blue dye at different concentrations: (a) channel at λ = 630 nm; and(b) channel at λ = 690 nm.

As expected from the previous results, the curves show an excellent linearity and have a slope thatdecreases with the increasing concentration of the dye. The slope is therefore consistent with the degreeof the darkness of the sample. This proves the capability of the device to discriminate between darksamples, which are apparently similar in a visible sense, by detecting the correspondent differences inspectral reflectance. This is the typical situation of dark fruits, like berries of red grape or blueberries.

On the basis of this and previous considerations, the solutions at different concentrations of dye couldbe used as handmade and cost-effective references to calibrate the instrument for measurements on darksamples. Regarding this possibility, the figure shows the horizontal threshold line and the operation point(see the diamond markers), which should be identified by calibrating the instrument with the solutions.As shown, the use of these kinds of handmade references makes it possible to set high values for theStimulus parameter, otherwise unreachable, particularly suitable for measuring dark fruits. A wider andmore detailed range of references could be realized according to a simple protocol and then replicatedwhen needed, depending on the specific application.

Sensors 2015, 15 22720

Table 3. Validation measurements on standard water solutions of blue dye at concentrationsof 0.5% (w/v), 1% (w/v) and 2% (w/v), respectively. Measurements on five repetitions in theform of Average Intensity ± Standard Deviation are presented. Abbreviations used in thetable: Calib. = Calibration; Std. = Standard.

ChannelReference Used in Calibration

Calib. Setpoint Measurements (Intensity [a.u.]) on Std. Solutions atλ[nm] Stimulus [a.u.] 0.5% (w/v) 1% (w/v) 2% (w/v)

630 Std. solution, blue dye 1% (w/v) 77 19, 433± 653 16, 577± 369 14, 273± 447

690 Std. solution, blue dye 1% (w/v) 103 19, 636± 719 16, 447± 387 14, 196± 519

750 Std. reference, R = 0.25 23 11, 741± 276 12, 996± 312 13, 725± 389

850 Std. reference, R = 0.50 14 7962± 231 8945± 289 10, 966± 300

The results coming from the final validation tests of the system in its operative configuration are shownin Table 3. To take into account measuring dark fruits as a possible application, the 1% (w/v) standardsolution was used as a reference for calibrating both visible channels (λ = 630, 690 nm). Conversely, inthe near-infrared spectral bands, the calibration was conducted with the standard reference R = 0.25 forthe channel λ = 750 nm, and the standard reference R = 0.50 for the channel λ = 850 nm, respectively.As desired, the calibration procedure consistently returned a different value for the Stimulus parameter,specific for every channel.

For the three measured samples, i.e., the three water standard solutions, the average measuredintensities and the standard deviations on five different acquisitions are reported. The data again areconsistent with the degree of the darkness of the acquired standard solution, confirming the capabilitiesof the device for discriminating very low levels of reflectance. These tests also prove the measurementrepeatability of the device system: indeed, the standard deviations are proportional to the measuredintensity and in general limited to 2%–4% of the measured value.

4. Conclusions

In this work, a prototype of a simplified optical VIS/NIR device based on LED technology wasdesigned and tested. Laboratory tests were conducted for the study of the characteristic curve ofthe instrument. In particular, the system response as a function of the two setting parameters wasinvestigated: the Gain amplification factor and the Stimulus level for the drive signal of the LEDs.In both cases, a highly linear response was obtained.

The optimal setting value for the amplification parameter has been determined as Gain = 1. Thisis accompanied by a semi-automatic calibration procedure implemented to allow the user to definean optimal Stimulus level for each LED driving signal independently. The system can therefore emita custom four-wavelength spectrum, optimized for the specific application. In this way and thanks tothe modularity given by the LEDs/filters units, the same device could be adapted for measuring differentproducts and for a wide range of applications.

The sensitivity of the device has highlighted the need for additional references. A simple andeconomic solution could be the definition of a protocol for the preparation of custom references. Inthis work, water solutions of standard food dyes at different concentrations were proposed.

Sensors 2015, 15 22721

Final validation tests on these kind of solutions prove the measurement repeatability of the devicewith a standard deviation limited to 2%–4% and confirmed its capabilities of discriminating low levelsof reflectance. This could be particularly useful when measuring dark fruits, like berries of red grapeor blueberries.

The integration of a simple processing algorithm in the microcontroller software would allow one tovisualize real-time values of reflectance. An evolution and an engineering phase is desirable in order toobtain a handheld, user-friendly and inexpensive device for farmers’ utilization in the field.

Author Contributions

The work was realized with the collaboration of all of the authors. Raffaele Civelli, ValentinaGiovenzana and Roberto Beghi contributed to the study design, collection, analysis and interpretationof the data; they also drafted the early version of the manuscript. Ezio Naldi contributed to the designand construction of the hardware and the prototype. Riccardo Guidetti organized the work, acquiredthe funding and supervised the research. Roberto Oberti contributed to the definition of the technicalspecifications and to designing the device and critically reviewed the draft of the paper. All authorsdiscussed the results and implications, commented on the manuscript at all stages and approved thefinal version.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix

A. Acquisition File on Standard Reference R = 0.99

Acquisition .txt file for a measurement on the standard reference R = 0.99, when using the instrumentin the explorative configuration mode for the evaluation of its characteristic curve.

B. Acquisition File on Standard Reference R = 0.50

Acquisition .txt file for a measurement on the standard reference R = 0.50, when using the instrumentin the explorative configuration mode for the evaluation of its characteristic curve.

C. Acquisition File on Standard Reference R = 0.25

Acquisition .txt file for a measurement on the standard reference R = 0.25, when using the instrumentin the explorative configuration mode for the evaluation of its characteristic curve.

D. Acquisition File on Standard Reference R = 0.02

Acquisition .txt file for a measurement on the standard reference R = 0.02, when using the instrumentin the explorative configuration mode for the evaluation of its characteristic curve.

Sensors 2015, 15 22722

E. Acquisition File on the Standard Solution at 0.5% (w/v)

Acquisition .txt file for a measurement on the aqueous solution of blue dye at a concentration of2% (w/v), when using the instrument in the explorative configuration mode for the evaluation of itscharacteristic curve at Gain = 1.

F. Acquisition File on the Standard Solution at 0.5% (w/v)

Acquisition .txt file for a measurement on the aqueous solution of blue dye at a concentration of1% (w/v), when using the instrument in the explorative configuration mode for the evaluation of itscharacteristic curve at Gain = 1.

G. Acquisition File on the Standard Solution at 0.5% (w/v)

Acquisition .txt file for a measurement on the aqueous solution of blue dye at a concentration of0.5% (w/v), when using the instrument in the explorative configuration mode for the evaluation of itscharacteristic curve at Gain = 1.

References

1. El-Masry, G.; Nassar, A.; Wang, N.; Vigneault, C. Spectral methods for measuring quality changesof fresh fruits and vegetables. Stewart Postharvest Rev. 2008, 4, 1–13.

2. Nicolai, B.M.; Beullens, K.; Bobelyn, E.; Peirs, A.; Saeys, W.; Theron, K.I.; Lammertyn, J.Nondestructive measurement of fruit and vegetable quality by means of NIR spectroscopy:A review. Postharvest Biol. Technol. 2007, 46, 99–118.

3. Wang, W.; Paliwal, J. Near-infrared spectroscopy and imaging in food quality and safety.Sens. Instrumen. Food Qual. 2007, 1, 193–207.

4. Guidetti, R.; Beghi, R.; Giovenzana, V. Chemometrics in food technology. In Chemometrics inPractical Applications; InTech: Rijeka, Croatia, 2012; pp. 217–252.

5. McClure, W.F.; Tsuchikawa, S. Instruments. In Near-infrared Spectroscopy in Food Science andTechnology; Yukihiro Ozaki, W., Fred McClure, A.A.C., Eds.; Wiley-Interscience: Hoboken, NJ,USA, 2007; Chapter 4.1, pp. 75–107.

6. Osborne, B.G.; Fearn, T.; Hindle, P.T. Practical NIR Spectroscopy with Applications in Food andBeverage Analysis, 2nd ed.; Longman Food Technology, Longman Scientific & Technical: Harlow,Essex, UK, 1993.

7. Antonucci, F.; Pallottino, F.; Paglia, G.; Palma, A.; D’Aquino, S.; Menesatti, P. Non-DestructiveEstimation of Mandarin Maturity Status through Portable VIS-NIR Spectrophotometer.Food Bioprocess Technol. 2011, 4, 809–813.

8. Camps, C.; Christen, D. Non-destructive assessment of apricot fruit quality by portable visible-nearinfrared spectroscopy. LWT - Food Sci. Technol. 2009, 42, 1125–1131.

9. Beghi, R.; Spinardi, A.; Bodria, L.; Mignani, I.; Guidetti, R. Apples Nutraceutic PropertiesEvaluation Through a Visible and Near-Infrared Portable System. Food Bioprocess Technol. 2013,6, 2547–2554.

Sensors 2015, 15 22723

10. Guidetti, R.; Beghi, R.; Bodria, L. Evaluation of Grape Quality Parameters by a Simple Vis/NIRSystem. Trans. ASABE 2010, 53, 477–484.

11. Bodria, L.; Fiala, M.; Guidetti, R.; Oberti, R. Optical techniques to estimate the ripeness ofred-pigmented fruits. Trans. ASABE 2004, 47, 815–820.

12. Guidetti, R.; Beghi, R.; Bodria, L.; Spinardi, A.; Mignani, I.; Folini, L. Prediction of blueberry(Vaccinium corymbosum) ripeness by a portable Vis-NIR device. Acta Hortic. (ISHS) 2009,810, 877–886.

13. Saranwong, S.; Sornsrivichai, J.; Kawano, S. Performance of a portable near infrared instrumentfor Brix value determination of intact mango fruit. J. Near Infrared Spectrosc. 2003, 11, 175–181.

14. Naes, T.; Isaksson, T.; Fearn, T.; Davies, T. A User-Friendly Guide to Multivariate Calibration andClassification; NIR Publications: Chichester, UK, 2002.

15. Wold, S.; Sjostrom, M.; Eriksson, L. PLS-regression: A basic tool of chemometrics.Chemom. Intell. Lab. Syst. 2001, 58, 109–130.

16. Cen, H.; He, Y. Theory and application of near infrared reflectance spectroscopy in determinationof food quality. Trends Food Sci. Technol. 2007, 18, 72–83.

17. Zou, X.; Zhao, J.; Povey, M.J.; Holmes, M.; Mao, H. Variables selection methods in near-infraredspectroscopy. Anal. Chim. Acta 2010, 667, 14–32.

18. Beghi, R.; Giovenzana, V.; Civelli, R.; Malegori, C.; Buratti, S.; Guidetti, R. Setting-upof a simplified handheld optical device for decay detection in fresh-cut Valerianella locusta L.J. Food Eng. 2014, 127, 10–15.

19. Giovenzana, V.; Beghi, R.; Malegori, C.; Civelli, R.; Guidetti, R. Wavelength Selection with a Viewto a Simplified Handheld Optical System to Estimate Grape Ripeness. Am. J. Enol. Vitic. 2014,65, 117–123.

20. Beghi, R.; Giovenzana, V.; Spinardi, A.; Guidetti, R.; Bodria, L.; Oberti, R. Derivation ofa blueberry ripeness index with a view to a low-cost, handheld optical sensing device for supportingharvest decisions. Trans. ASABE 2013, 56, 1551–1559.

21. Costa, G.; Bonora, E.; Fiori, G.; Noferini, M. Innovative non-destructive device for fruit qualityassessment. Acta Hortic. (ISHS) 2011, 913, 575–581.

22. Wu, D.; Feng, L.; Zhang, C.; He, Y. Early detection of Botrytis cinerea on eggplant leaves basedon visible and near-infrared spectroscopy. Trans. ASABE 2008, 51, 1133–1139.

23. Oberti, R.; Marchi, M.; Tirelli, P.; Calcante, A.; Iriti, M.; Borghese, A.N. Automatic detection ofpowdery mildew on grapevine leaves by image analysis: Optimal view-angle range to increase thesensitivity. Comput. Electron. Agric. 2014, 104, 1–8.

c© 2015 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access articledistributed under the terms and conditions of the Creative Commons Attribution license(http://creativecommons.org/licenses/by/4.0/).